by Daniel Sutton
In 1960, three years before the first cassette tape was invented, a group of Japanese architects calling themselves ‘Metabolists’ produced a manifesto, which outlined a series of goals for society. In one section, they made an especially bold prediction: ‘Everyone will have a brain wave receiver’ in his ear, which conveys directly and exactly what other people think about him and vice versa...There is no more individual consciousness, only the will of mankind as a whole.’ That same decade, UCLA academic F.M. Esfandiary would rename himself FM-2030 and begin lecturing on the advent of ‘Transhumanism’, computer scientist Marvin Minsky would outline the complementary roles of AI and the human mind, and Iron Man would enter the Marvel Universe.
Fifty-seven years on, these same ideas are enjoying a renaissance. On cinema screens, Scarlett Johansson’s cybernetically-enhanced Mira has drawn $160 million in box-office receipts. In the Italian Parliament, Guiseppe Vatinno sits as the first politician elected with a Transhumanist manifesto. This time, however, imagination is being matched by innovation. At the time of writing, billionaire Tesla CEO Elon Musk is assembling a team of elite scientists in San Francisco for his newest company, Neuralink, with a pledge to ‘develop ultra-high bandwidth brain-machine interfaces to connect brains and computers’. Nicknamed the ‘Neural Lace’, this is the Metabolist’s vision is on the cusp of reality.
For all its imagination, however, Musk’s vision is rather one-dimensional. Of Neuralink’s founding board, only one member has come from outside the highest echelons of neuroscience and engineering: Musk himself. The company is advertising to recruit scientists who specialize in microfabrication, brain-machine interfacing and immunohistochemistry, but not a single role asking for experience in man over machine. Musk has repeatedly assumed that cybernetic enhancements, properly constructed, can only be beneficial, and that artificial intelligence holds the capacity to outstrip the human brain in all meaningful regards. In his most recent interview in Dubai, he described the only reason for balance between human and machine intelligence as a trade-off between ‘control’ and ‘usefulness’. He has made no mention of the creative or empathetic abilities of the human brain, instead judging it by the same criteria as one might a processor. This perspective is a recent development – note how the Metabolists hoped to pool all mankind’s intelligence, while Musk would prefer to pair each brain with a machine – but has widespread currency in popular culture. Down the Californian coast at the Burbank studios, Marvel’s superhero Vision serves as a symbol of the unilateral superiority of artificial intelligence; on the day Neuralink launched, meanwhile, shares in Tesla soared.
Are the days of the human brain numbered, then? To find someone looking to argue the opposite, asking whether human intelligence has unique abilities computers could never surpass, we must return to Oxford. While Musk draws the brightest young talent from across America, 85 year-old Professor of Mathematics Sir Roger Penrose has begun investigating whether there are some things the human brain can do that no machine ever could. His theory is that quantum effects in the brain allow humans to perform calculations far beyond any machine’s capability – we understand this as ‘creative’ intelligence. He still accepts Musk’s premise that all useful brain functions are calculations, but argues our brain has unique capabilities in that regard. Currently, the Penrose Institute is trying to prove this hypothesis by constructing puzzles which humans alone can solve, or at least solve much faster than computers. By analysing participants’ neural patterns as they tackle the puzzles, Penrose hopes to explain the source of human creativity or intuition.
Earlier this year, Penrose publicly released his first puzzle, the chess problem pictured below. The subject matter was pointedly chosen, as supercomputers now easily beat the world’s strongest chess players, and computer-assisted play is growing in popularity. Penrose claimed that while many computers, trying to solve the problem move-by-move, would assume that Black was winning because he had far better pieces, a moderately skilled human would quickly see it was a draw through principle, because Black cannot make progress. Penrose also hoped that many more powerful computers would struggle to solve the puzzle because they would take a very long time to find such an odd position in their databases, while the human is unfazed by originality.
The results of Penrose’s problem are not yet known, but it seems likely it has broadly worked as intended. When I put the problem to the best chess computer in the world, Komodo 10 (64 bit), and a range of members of Oxford University Chess Club (all strong amateurs), Komodo assumed that Black was easily winning, while all of the humans reached the correct solution relatively swiftly. In the course of my survey, however, there were a couple of unexpected twists.
First, Penrose’s position is unique. Though there are a handful of other known positions where humans have found solutions computers cannot, only the very strongest players in the world have even a chance of solving them, and typically do so with computer assistance. Considering that the number of legal chess positions is 10121, the human achievement looks very small indeed. On this basis, if we were designing a cybernetically-engineered human for the sole purpose of playing chess, we might as well make the entire brain artificial: even if there are odd cases where humans outperform the machine, the probability of them showing up is so low that anything else would be inefficient.
One conclusion we could draw is that human intelligence is, practically-speaking, doomed to inferiority, and Penrose’s exception is small comfort. Most of the chess-players I surveyed held this view, and argued that Komodo could easily be given the extra information to solve the position if that were important. Musk also seems to be an adherent of this view, focusing on the need to retain control and accept human inabilities. The other possibility, however, is that Penrose’s position is an anomaly because it draws on parts of the brain chess rarely uses, and what he has really shown is not that human intelligence has useful superiorities, but that chess is limited as a context for that test.
This idea finds support in a second unexpected finding. The puzzle is perhaps not as uniformly easy for humans as Penrose thinks: while his grandmaster brother might have had no problem with it, it took some of the surveyed chess-players a few minutes thought, while others managed it in seconds. There was no correlation between chess-playing standard and the speed of solution; when the participants were asked about how they approached the puzzle, however, an interesting pattern emerged. The faster players did not, as he expected, solve it through abstraction – working out that it was a draw through the principle neither side could make progress – but by instinctively asking why Penrose had set it up like he had (specifically, why most of the Black pieces were bunched on that side and there were three dark-squared bishops). The slower ones were those who came to this method late, or used move-by-move reasoning (like a computer) to reach a principle.
Arguably, the quicker players had not played by the spirit of the test, because the computer never had access to the information about the position’s context, and in a proper match such a tactic would never work. It does suggest, however, that the advantage the human had in Penrose’s puzzle was not calculating ability, as Penrose expected, but empathetic ability (understanding someone else’s perspective from experience). The faster the humans imagined the position from Penrose’s perspective, the quicker they saw what he was trying to prove. Having the most powerful computer in the world linked to their brain would not have helped them at all. On a very basic level, this conclusion challenges an assumption held by both Penrose and Musk: that intellectual capability (or as Musk put it, ‘usefulness’) is measurable as the speed and accuracy with which one can calculate a solution to a given problem. The most efficient solution appears to be based on finding a new problem with the same solution, but from a different perspective: something that, for those who were willing to abandon conventional chess practice, was not just easy, but instinctive.
Musk and Penrose are hardly alone in understanding the brain as a calculating mechanism. Even our moral decisions, although typically made intuitively, are often expressed in quantitative terms: a utilitarian calculus, Pareto superior outcomes, distributive justice. But this model does not reflect the nature of human ingenuity. When Penrose set out to create his puzzle, he cannot have worked through the 10121 positions in any logical order: he surely gathered enough information to imagine how a supercomputer would try to solve a given puzzle, then matched up features it would typically struggle with. He demonstrated the same attributes as the faster players at Oxford University Chess Club: imagining from another perspective, and trying to rewrite the rules.
Although it might not work in the manner he intended, Penrose’s position shows that there is plenty of the human brain it would be useful to retain, even if usefulness were the only reason to do so. It suggests that Musk and his team have mistaken man for an outmoded machine, and forgotten how they communicate their visions, design software that sells and raise public awareness: by imagining the result from someone else’s perspective. Whether artificial intelligence could ever match this ability is currently unclear. Earlier this year, computer Libratus, defeated a range of world-leading poker players, demonstrating the ability to anticipate a human bluff. But it did so through a blend of calculation and machine learning, and with imperfect results. At the beginning of May, the first ‘emotional chat-bot’ was launched in China. Learning emotion by studying thousands of Weibo posts, it can adjust its tone according to human responses, but still fails to pass the Turing Test or detect skilful lies. Artificial intelligence still has a significant leap to make before it can match even the effects of human empathy and imagination.
The Metabolists, just like many of their contemporaries, displayed a brash optimism: they assumed that science and society would evolve hand in hand, and that their duty was to shape that growth as best they could. Fast- forward to today, and it is fear that underpins cybernetic visions. As Elon Musk most memorably put it, a neural lace is necessary to prevent us becoming a machine’s ‘house cats’. Musk may be right about the powerful potential of artificial intelligence, and his crack team of neuroscientists and engineers may well revolutionise that field. But on the other side of the Atlantic, Penrose’s chess puzzle seems to show that the human mind has plenty more to offer before it becomes a pet, with empathetic ability at its heart. Indeed, it shows that humans still have the capacity to both devise and solve problems which outsmart a world-leading machine. As ever in Oxford, tradition is not ready to move on yet.
DANIEL SUTTON reads Ancient and Modern History at St John's. He likes a steady diet of Classical literature and chocolate cookies.